Upload
vunguyet
View
229
Download
5
Embed Size (px)
Citation preview
The Labor Market and Macro Volatility: A NonstationaryGeneral-Equilibrium Analysis∗
Robert E. HallHoover Institution and Department of Economics
Stanford UniversityNational Bureau of Economic Research
stanford.edu/∼rehall
September 22, 2005
Abstract
The evolution of the aggregate labor market is far from smooth. I investigate thesuccess of a macro model in replicating the observed levels of volatility of unemploy-ment and other key variables. I take variations in productivity growth and in exogenousproduct demand (government purchases plus net exports) as the primary exogenoussources of fluctuations. The macro model embodies new ideas about the labor market,all based on equilibrium—the models I consider do not rest on inefficiency in the useof labor caused by an inappropriate wage. I find that non-standard features of the labormarket are essential for understanding the volatility of unemployment. These modelsinclude simple equilibrium wage stickiness, where the sticky wage is an equilibriumselection rule. A second model based on modern bargaining theory delivers a differentkind of stickiness and has a unique equilibrium. A third model posits fluctuations inmatching efficiency that may arise from variations over time in the information aboutprospective jobs among job-seekers. Reasonable calibrations of each of the three mod-els match the observed volatility of unemployment.
∗I am grateful to Felix Reichling for comments. This research is part of the program on Economic Fluc-tuations and Growth of the NBER. A file containing data and programs is available at Stanford.edu/∼rehall
1
1 Introduction
Observed movements of employment and unemployment are larger than standard general-
equilibrium models can comprehend, given the relatively small exogenous shocks that hit
the economy. I investigate this topic in a family of models with labor markets that contain
amplification mechanisms that may contribute to the understanding of volatility.
The models incorporate some standard features of modern general-equilibrium macroe-
conomics. Households plan consumption according to a standard Euler equation. Invest-
ment faces adjustment cost. With respect to the labor market, I depart from the main
tradition of real business cycle and dynamic general equilibrium macro models. That tradi-
tion has considered the margin between employment in the market and time spent at home
and has neglected time spent looking for work. I follow the other important tradition of
modern macroeconomics, the matching model of job search. This line of thought considers
the margin between working and looking for work and takes labor-force participation to be
inelastic. In place of the Nash-bargain wage determination usually considered in that tra-
dition, I consider two alternatives. First is the equilibrium sticky-wage formulation in Hall
(2005c). The wage is less responsive to current conditions, though not so sticky as to cre-
ate bilaterally inefficient outcomes between employer and worker. Second is the bargaining
model of Hall and Milgrom (2005), which replaces non-credible threats to abandon wage
bargaining with credible threats to extend bargaining. That model achieves a sticky-wage
result by limiting the role of conditions in the labor market in wage determination. I also
consider a model where fluctuations in matching efficiency drive unemployment fluctua-
tions. The last model mimics fluctuations arising from changes in the quality of information
available to job-seekers that induce alterations in self-selection, following Hall (2005a).
The economy in this paper is explicitly non-stationary. I present essentially exact nu-
merical solutions to a stochastic growth model whose levels of output, consumption, and
capital have unit roots, inherited from the unit root in the efficiency of production. Rather
than deal with filtered data to remove the unit root, I state the key variables as ratios to the
2
capital stock. The model is stationary in these ratios.
I measure volatility in terms of changes in the variables. This approach captures the
cyclical component of volatility along with the important movements that are not necessar-
ily associated with any concept of a business cycle. The model and associated measurement
approach do not rest on any attempt to separate cyclical movements from other movements.
This paper is a further development of the efforts of Merz (1995), Andolfatto (1996),
and Alexopoulos (2004) to incorporate unemployment in general-equilibrium macro mod-
eling. The most important difference from earlier work is in wage determination and
the amplification of fluctuations that occurs with equilibrium sticky wages and other non-
standard features of the labor market.
I find that departing from the standard view of the labor market is essential for under-
standing the volatility of unemployment. As Shimer (2005) has demonstrated, the standard
Mortensen and Pissarides (1994) setup cannot rationalize movements of unemployment of
the observed magnitude as the result of productivity shocks of the observed magnitude.
Non-standard labor-market models provide the needed amplification mechanisms. A basic
idea common to all of the models—discussed in Hall (2005c)—is that employers deter-
mine the level of recruiting effort based on the expected share of the joint value created
when they form matches with workers. In times when the share is low, recruiting effort
is correspondingly low. Job-finding rates for the unemployed are low and the unemploy-
ment rate is high. Matches are formed and retained efficiently, according to the principle
of maximizing joint value. Bilateral efficiency implies that the model’s view of the labor
market is one of economic equilibrium, not the disequilibrium previously associated with
amplification mechanisms based on sticky wages.
The model overcomes one of the most persuasive criticisms of real business-cycle
models—that the models portray recessions as the result of actual contractions in produc-
tivity (see Summers (1986)). Here, in a growing economy, a recession with abnormal un-
employment will occur when productivity growth is positive but lower than normal. I also
3
show that movements in exogenous product demand are important sources of fluctuations.
This driving force has received less attention from earlier models.
At reasonable values of the key parameters—intertemporal elasticity of substitution,
capital adjustment cost, and wage stickiness—the models account reasonably well for the
volatilities of consumption growth, the investment/capital ratio, the capacity/capital ratio,
and unemployment.
2 Model
2.1 Shocks
Serially independent shocksat affect productivityAt, which evolves as
At+1 = atAt. (1)
The shocks are drawn from a discrete set,{a1, ..., a5} with probabilities{ψ1, ..., ψ5}. Se-
rially correlated shocksgt affect exogenous product demand. They are stated as ratios to
consumption,ct, so the shocks to product demand aregtct. They are drawn from a discrete
set{g1, g2, g3} and obey a first-order Markoff process:
Prob[gt+1 = g′|gt = g] = πg,g′ (2)
2.2 Technology, capital, and consumption
Let yt be output produced at timet, At be the efficiency of production,nt be employment,
kt be capital,xt = kt+1/kt be the capital growth ratio (also the investment/capital ratio),
vt be resources expended in recruiting workers, andct be consumption.τ is a parameter
controlling adjustment cost. Labor and capital form gross output according to
Atnγt k
1−γt . (3)
Adjustment costs are:1− τ
2τkt(xt − 1)2. (4)
4
The government purchases part of output and another part leaves the country as net exports.
I take the sum of the two to be an exogenous fraction,gt, of consumption. Because the
model is nonstationary, the exogenous component of product demand needs to be linked to
a variable that grows according to the economy’s stochastic trend; consumption is the most
suitable choice because it is the most stable.
Capital deteriorates at rateδ, so it follows the law of motion
kt+1 = (1− δ)kt + Atnγt k
1−γt − 1− τ
2τkt(xt − 1)2 − vt − (1 + gt)ct. (5)
I let qt be the market price of installed capital att. Firms solve the atemporal capital-
installation problem:
maxxt
qtxtkt − 1− τ
2τkt (xt − 1)2 − xtkt. (6)
The first-order condition is:
xt =τ
1− τ(qt − 1) + 1, (7)
Tobin’s investment equation. The Tobin coefficient,τ , controls capital adjustment. Ifτ =
0, capital does not adjust at all; the economy is the endowment economy of Lucas (1978).
If τ = 1, capital adjusts without impediment andq is always one.
Households can buy and sell a claim to a unit of installed capital with priceqt. Its return
ratio is
Rt+1 =(1− δ)qt+1 + (1− γ)At+1n
γt+1k
−γt+1
qt
. (8)
Households have an intertemporal elasticity of substitution ofσ. Their intertemporal marginal
rate of substitution is
mt+1 = β(
ct+1
ct
)− 1σ
. (9)
The marginal rate of substitution prices all economic values in periodt + 1. I define the
present value operator,
Pt(Xt+1) = Et (mt+1Xt+1) . (10)
5
Households plan consumption to satisfy the Euler equation,
Pt (Rt+1) = 1. (11)
This setup makes the implicit assumption that individuals are fully insured for the idiosyn-
cratic risk of unemployment—all individuals consume the same amount, independent of
their employment situations. See Merz (1995) for further discussion of this point.
2.3 Labor market
The labor market operates according to principles laid out in Hall (2005c), Hall (2005b),
and, in a later section, Hall and Milgrom (2005), based on some of the features of Mortensen
and Pissarides (1994).
I normalize the labor force at one. The number of workers who lose their jobs each
quarter is a constant,s. Among the job-losers, a fractionf find new jobs immediately. The
remainder find jobs after a quarter. Thus the unemployment rate isut = s(1 − ft). The
separation rate—the fraction of workers who lose their jobs—is
st =s
1− ut
. (12)
This setup mimics actual unemployment reasonably well and has the (substantial) modeling
benefit that the unemployment rate is not a state variable. The emphasis on the probability
that a job-loser becomes unemployed is realistic—much of the variation in U.S. unemploy-
ment arises from fluctuations in the fraction of job-losers who become unemployed—see
Hall (2005d).
Job-seekers receive unemployment benefits at a flow rateλt, financed by a lump-sum
tax.
Workers receive compensation with present valueWt from jobs starting in quartert.
The Bellman value,Ut, associated with being unemployed is the sum of expected unem-
ployment benefits, the wage,Wt+1, paid by the job to be found next quarter, and the value,
6
Vt+1, of being employed at that time, apart from the wages on that job:
Ut = λt + Pt(Wt+1 + Vt+1). (13)
The worker may (1) separate next quarter and find another job immediately, (2) separate
next quarter and enter unemployment, or (3) remain at work.Vt is the present value, while
employed, of these three possible states next quarter:
Vt = Pt (st+1ft+1(Wt+1 + Vt+1) + st+1(1− ft+1)Ut+1 + (1− st+1)Vt+1) . (14)
I let Zt be the present value of the marginal product of the worker on a particular job,
at the time the job begins. It satisfies the recursion,
Zt = γAt(1− ut)γ−1k1−γ
t + Pt ((1− st+1)Zt+1) . (15)
Employers devote an amount of output,vt, to recruiting and qualifying workers—vt
includes all expenditures on workers that occur before the terms of a match are set through
a wage bargain. Employers’ recruiting spending determines the job-finding rate,ft. When
employers are looking actively for new workers, jobs are easy to find andft is high. The
marginal benefit of added recruitment spending in terms of the job-finding rate declines
with vt according to a concave functionφ(vt/kt). I scale by capital because recruiting and
qualifying workers is more expensive in a more advanced economy.
My treatment of the matching process departs from the standard in the matching lit-
erature for reasons of modeling convenience, not substance. Linking the job-finding rate
directly to recruiting expenditure can be seen as a reduced form of the usual set-up, where
vacancies appear explicitly—see Hall (2005b).
Employers earn a flow of profit equal to the benefit from hiring one new worker (the
employer’s share of the surplus for a job,Zt−Wt) multiplied by the flow of new hires,sft,
less the cost of recruiting the flow of workers,vt. Employers compete for the services of
job-seekers up to the point of zero profit. The zero-profit condition is
vt = (Zt −Wt)sft. (16)
7
2.4 Stationary model
Before discussing wage determination, I create a stationary version of the model by re-
defining the following variables to be their earlier values divided by the capital stock,kt:
consumption,ct, recruiting resources,vt, government purchases,gt, and the four labor-
market values,Ut, Vt, Zt, andWt. I also introduce the state variablezt = Atk−γt , the ratio
of capacity,Atk1−γt , to capital,kt. In the labor market, I assume that unemployment com-
pensation is a constant fraction of the marginal product of labor:λt = λγzt(1− ut)γ−1kt
The model becomes
zt+1 = at+1x−γt zt (17)
xt = 1− δ + zt (1− ut)γ − 1− τ
2τ(xt − 1)2 − vt − (1 + gt)ct (18)
xt =τ
1− τ(qt − 1) + 1 (19)
Rt+1 =(1− δ)qt+1 + (1− γ)zt+1(1− ut+1)
γ
qt
(20)
mt+1 = β(xt
ct+1
ct
)− 1σ
(21)
Pt (Rt+1) = 1 (22)
Ut = λγzt(1− ut)γ−1 + Pt(Wt+1 + Vt+1)xt (23)
Vt = Pt (st+1ft+1(Wt+1 + Vt+1) + st+1(1− ft+1)Ut+1 + (1− st+1)Vt+1) xt (24)
Zt = γzt(1− ut)γ−1 + Pt ((1− st+1)Zt+1) xt (25)
vt = (Zt −Wt)sft (26)
ft = φ(vt) (27)
ut = s(1− ft) (28)
st =s
1− ut
. (29)
8
2.5 Wages
I will consider the following model of wage determination:
Wt = κW ∗ + (1− κ)1
2(Ut − Vt + Zt). (30)
The quantity12(Ut − Vt + Zt) is the average of the reservation wage of the jobseeker,
Ut − Vt, and the reservation wage of the employer,Zt. It is the wage set in a symmetric
Nash bargain, as in the earlier literature. To the extent thatκ > 0, the wage is sticky
compared to the Nash-bargain wage, as in Hall (2005c). It does not respond as much to the
immediate ups and downs of productivity. Ifκ = 0, the Nash bargain sets the wage. This
wage rises when a favorable productivity shock,at, hits the economy. Output rises more
than the capital stock—this is true without capital adjustment costs, but is even more true in
their presence. The scarcity raises the marginal product of labor, which is proportional to
the state variablezt—see equation (25). The actual wageWt responds immediately through
the Nash bargain. As capital rises to restore the normal output/capital ratio,zt returns to its
normal value and the wage is back to normal in relation to the capital stock.
I set the parameterW ∗ to the level of the wage/capital ratio in the stationary non-
stochastic Nash-bargain model—the normal wage in the flexible-wage model. In the fixed-
wage model (κ = 1), the wage/capital ratio,Wt, is constant at the levelW ∗. That is, the
wage remains at its normal level in relation to the capital stock rather than rising when
productivity growth is strong and falling when it is weak.
Intermediate values ofκ describe an economy with partial wage stickiness.
Equation (30) may place the wage outside the bargaining set—that is, below the job-
seeker’s reservation wageUt − Vt or above the employer’s reservation wage,Zt. I mod-
ify the equation to keep the wage within the bargaining set. Thus the sticky wage lies
within the equilibrium set described in Hall (2005c). Sticky wages never cause inefficient
separations—they cause unemployment volatility through their influence on employers’
recruiting efforts.
9
2.6 Other issues
I assume that the productivity and exogenous spending shocks each take on one of a finite
number of possible values. In this setting, the expectations in the Euler equations are
summations rather than integrals.
To solve the model, I represent each endogenous variable as a function of the state vari-
ables,z andg, with a finite set of parameters (specifically, the coefficients of a Tchebysheff
polynomial of order 9). I write the model in terms of the current value ofz and its value one
year in the future,z′ = ax−γz. The resulting model is a large system of nonlinear equa-
tions in the coefficients of the functions, which can be solved by standard methods. Based
on the solution, I represent the entire model as a first-order Markoff process of very high
dimension. I calculate the stationary distribution from the Markoff matrix and derive the
(essentially) exact standard deviations of the endogenous variables from that distribution.
For further details, see Judd (1998) and the appendix to this paper.
Notice that the model embodies a transversality condition that the variables tend toward
a stationary point. In other words, a solution corresponds to a terminal condition in the
distant future, not a condition that has an appreciable effect on the economy today. The
transversality condition takes the form of the requirement that the consumption rulecg(z)
depends only onz andg. If the economy faced a terminal condition in finite time, the
consumption rule would depend on that condition as well.
3 Volatility Measures
Table 1 shows data for the years 1949 through 2000 for the volatility measures that I con-
sider. The first is productivity growth, taken as an exogenous driving force of fluctuations.
The standard deviation of quarterly growth in total factor productivity is 0.94 percent. The
second is the ratio of exogenous spending to consumption, another driving force. Its stan-
dard deviation is 4.8 percent.
10
Variable Data source Units Value
Productivity growth Bureau of Labor Statistics, Multi-
factor Productivity Index, Private
Business
Quarterly standard
deviation, percent,
inferred as half the
annual standard
deviation
0.94
Government purchases plus
net exports relative to
consumption
Bureau of Economic Analysis,
National Income and Product
Accounts, Table 1.1.5. Nominal
Gross Domestic Product
Standard deviation
of quarterly data,
percent
4.83
Consumption growth Bureau of Economic Analysis,
National Income and Product
Accounts,Table 1.1.3. Real Gross
Domestic Product, Quantity Indexes
Standard deviation
of quarterly data,
percent
0.88
Investment/capital ratio Growth ratio of the real capital stock
from Bureau of Economic Analysis,
Fixed Asset Tables, Table 1.2. Chain-
Type Quantity Indexes for Net Stock
of Fixed Assets and Consumer
Durable Goods
Standard deviation
of annual data,
divided by 4,
percent
0.22
Log of capacity/capital ratio Ratio of productivity index to (capital
stock raised to the power 0.7), with
constant exponential trend removed
Standard deviation
of log of annual
data, multiplied by
100
4.43
Unemployment Bureau of Labor Statistics,
Unemployment rate, 16 years and
older
Standard deviation
of quarterly data,
percent
1.57
Table 1. Data on macro volatility in the U.S., 1949-2000
11
The remaining measures are for variables taken to be endogenous in the model. Con-
sumption growth has a quarterly standard deviation of 0.88 percent. The standard deviation
of the investment/capital ratio (the growth rate of the capital stock) is 0.22 percent. I take
unemployment to be stationary, so I measure its volatility as the standard deviation of its
level, 1.57 percent.
4 Functional Forms and Parameter Values
In my discussion of the volatility implied by the model, I will start from a base case for
the key parameters: I take the wage stickiness coefficientκ to be 0.43, implying a wage
that responds only partially to the current state of the economy. I take the base value of the
intertemporal elasticity of substitutionσ to be 0.4, somewhat above the value suggested in
Hall (1988), in deference to suggestions in the recent literature of downward bias in my
estimates. The implied value of the coefficient of relative risk aversion is 2.5, a reasonable
value in terms of the discussion in Lucas (1994). Finally, I take the base value of the Tobin
coefficient for capital adjustmentτ to be 0.25—see Hall (2004).
I consider the following alternative values of the key parameters: Flexible wages, with
κ = 0, less intertemporal substitution, withσ = 0.2 (with higher coefficient of relative risk
aversion, 5), and faster capital adjustment, withτ = 0.5.
The driving forces for fluctuations in the model are technology growth,at, and ex-
ogenous spending relative to consumption,gt. In both cases, I convert time-series data to
discrete values by placing them in bins. I define the bins so that an equal fraction of the
observations fall in each bin. In the case of technology growth, I take the deviation from
the mean to be half the annual growth rate, to match the standard deviation of the observed
annual data. The average values for quarterly productivity growth in the five bins are -0.90,
-0.11, 0.26, 0.79, and 1.56 percent.
For exogenous product demand (the ratio of government purchases and net exports to
consumption), I work directly with quarterly data. Table 2 describes the distribution. The
12
Value 24.5 31.3 36.7
24.5 93.5 6.5 0.0
31.3 7.8 85.7 6.5
36.7 0.0 6.5 93.5
Transition probabilities from column value
to row value
Table 2. Transition frequencies for exogenous spending as a percent of consump-tion
average value ofgt in each bin are shown in the columns labeled “Value.” The table shows
the transition frequencies among the three categories.
I take the rate of deterioration of capital and the labor elasticity of the production func-
tion to have the standard valuesδ = 0.025, andγ = 0.7. I pick the utility discount so that
the average return to capital is 5 percent per year:β = 0.9992.
I take the recruiting technology to have the form
f = φ (v/k) = ω (v/k)1/2 . (31)
I pick the efficiency parameter of the recruiting technology,ω, by setting up a version of
the model with constant, non-stochastic productivity growth at the average rate from the
values in Table 2. I prescribe that the unemployment rate be its post-1949 average of 5.5
percent. I take the exogenous separation rates to have the value 10 percent per quarter per
month (see Hall (2005d) for a discussion of the measurement of the separation rate). Then
I solve the model comprising equations (18) through (30) plus an additional equation that
gives the condition for a stationaryz,
gx−γ = 1 (32)
The solution provides the valueω = 8.9.
13
5 Properties of the Model
The state variables,zt, the capacity/capital ratio, andgt, exogenous spending, capture the
current departure of the economy from its steady state. A favorable productivity shock
increases capacity immediately and raiseszt correspondingly. Investment responds imme-
diately and capital begins to rise, drivingzt back to its normal level until the next shock
occurs. Productivity shocks are permanent in the model. On the other hand, exogenous
spending shocks,gt, are persistent but ultimately die out, as ratios to consumption. A
strictly permanent spending shock would depress consumption permanently by the same
amount and leave all other variables unaffected. In the case where the spending shock dies
out over time, investment falls at the time of the shock. The cumulating decline in the
capital stock causeszt to rise gradually in the period after the spending shock. Investment
rises later to restore the steady-state value ofzt as spending returns to normal. Thuszt has
an exponential response to productivity impulses and a hump-shaped response to spending
impulses.
Table 3 reports the calculated volatilities for a variety of combinations of the key pa-
rameters. The base case, in the first column of results, fits the observed volatilities fairly
well. The model slightly overstates the volatilities of consumption growth and the invest-
ment ratio and matches the volatilities of the capacity/capital ratio and unemployment fairly
closely. Of course, the match is the result of adjusting parameter values, so Table 3 is really
the result of applying indirect inference informally. As I will show shortly, the exogenous
spending shock is an important contributor to the model’s ability to explain volatility—in
the presence of only the technology shock, the model understates volatility.
The second column of results shows the main point of the paper. Replacing the sticky-
wage specification with the standard Nash-bargain model of wage determination, I find
vastly too little volatility of unemployment. This finding confirms, in a general-equilibrium
setting, Shimer (2005)’s point that the standard matching model cannot generate anything
like a realistic account of the movements of unemployment. Notice that the volatilities of
14
Parameter Actual Base
Flexible
wage
Lower
inter-
temporal
substi-
tution
Lower
adjustment
cost
, intertemporal elasticity of
substitution0.4 0.4 0.2 0.4
, adjustment cost parameter 0.25 0.25 0.25 0.5
, wage stickiness parameter 0.43 0 0.43 0.43
Standard deviations
Consumption growth 0.88 0.94 0.98 1.03 0.90
Investment/capital ratio 0.22 0.22 0.23 0.28 0.23
Log of capacity/capital ratio 4.43 4.59 4.79 6.10 4.53
Unemployment 1.57 1.59 0.24 1.66 1.54
Table 3. Actual and calculated volatilities
15
the other variables are reasonably well matched. One must look in the labor market itself
to see that something important is missing from a general-equilibrium model. Standard
models can account for the volatilities of consumption growth and the other non-labor
variables.
The column of Table 3 headed “Lower intertemporal substitution” considers an econ-
omy with an intertemporal elasticity,σ, of 0.2 in place of the base value of 0.4. This
economy has higher consumption volatility because the consumption substitution effect is
smaller and offsets the wealth effect less. The volatilities of the investment/capital and ca-
pacity/capital ratios are—unrealistically—much higher because consumption absorbs less
of each shock.
The column of Table 3 headed “Lower adjustment cost” describes an economy with
a higher response of investment to Tobin’sq. The coefficient,τ , is 0.5 in this economy
against 0.25 in the base economy. A lower adjustment cost has remarkably little effect on
the volatilities implied by the model.
5.1 Endogenous variables as functions of state variables
An instructive way to display the model is in terms of functions such asug(z) that show
how the key variables depend on the two state variables. These functions, represented as
polynomials inz, are the results of solving the model, as described in the Appendix.
Figure 1 showsug(z) for the intermediate value of exogenous spending,g. The differ-
ence is dramatic between the response of unemployment in the sticky- and flexible-wage
models. The curve at the bottom of this and later figures shows the marginal probability
distribution of the capacity/capital ratio,z. In the sticky-wage model, the wage remains
close to its normal level in relation to the capital stock even when a disappointment in pro-
ductivity growth causes the capacity/capital ratio to drop below normal. In this case, the
wage is higher than the flexible wage and employers have correspondingly less incentive
to deploy resources in attracting new workers. The job-finding rate is lower than normal
16
0.00
0.01
0.02
0.03
0.04
0.05
0.06
0.07
0.08
0.09
0.10
0.115 0.120 0.125 0.130 0.135 0.140 0.145 0.150
Capacity/capital ratio
Un
em
plo
ym
en
t
Flexible wage
Sticky wage
Distribution
Figure 1. Unemployment responses, base case and flexible wage
and unemployment is high. The reverse holds for good news in productivity, which brings
forth high recruiting effort and low unemployment.
By contrast, Figure 1 shows that unemployment is barely responsive to shifts in the
capacity/capital ratio when wages are flexible. Adjustments of the wage result in only
small changes in the incentive to hire and corresponding unemployment rate. The figure
confirms, in a full dynamic general-equilibrium setting, the point of Shimer (2005) that the
Mortensen-Pissarides model with Nash-bargain wage setting cannot explain the volatility
of unemployment given the known amplitude of the driving force.
Figure 1 makes it clear that the flexible-wage model cannot come close to matching
the observed volatility of unemployment. From Table 1, unemployment has a standard
deviation of 1.57 percentage points. The model—with base-case parameters except for
full wage flexibility (κ = 0)—implies a standard deviation of unemployment of only 0.24
percentage points. Accordingly, I devote most of the attention in the discussion to the
17
0.990
0.995
1.000
1.005
1.010
1.015
1.020
0.115 0.120 0.125 0.130 0.135 0.140 0.145 0.150
Capacity/Capital Ratio
Inv
es
tme
nt/
Ca
pit
al
High
Medium
Low
Exogenous
spending
Distribution
Figure 2. Investment as a function of the state variables
sticky-wage case (κ = 0.43).
Figure 2 shows one of the key relationships in the model, the value of investment (mea-
sured by the growth factor for capital,x) as a function of the capacity/capital ratio,z, and
the level of exogenous spending,g. Investment responds to spending because spending
shocks, though highly persistent, are not permanent. The economy responds to spending
shocks in part by deferring investment until the time when the shock begins to subside.
Figure 3 shows consumption as a function of the capacity/capital ratioz and the discrete
state of exogenous spending. Because the spending shocks are quite persistent, consump-
tion responds significantly to the shocks.
The solution of the model includes a functionWg(z) that describes the wage as a func-
tion of the capacity/capital ratio,z, and the discrete state of exogenous spending,g. The
model requires that the wage lie within the bargaining set[U − V, Z]. The model solution
also supplies the functionsUg(z), Vg(z), andZg(z). Figure 4 shows the relation between
18
0.050
0.055
0.060
0.065
0.070
0.075
0.080
0.085
0.090
0.12 0.12 0.13 0.13 0.14 0.14 0.15 0.15
Capacity/Capital Ratio
Co
ns
um
pti
on
HighMedium
Low
Exogenous
spending
Distribution
Figure 3. Consumption as a function of the state variables
the actual wageW (z) and the boundaries of the bargaining set, for the intermediate value
of exogenous spending,g = g2. The actual wage reaches the upper boundary at the low-
est values ofz, but, as the distribution at the bottom of the figure shows, this outcome is
extremely rare. In bad times, employers gain hardly any of the surplus from a job match
and consequently put almost no resources into recruitment. The unemployment rate rises
to serious recession levels.
5.2 Impulse response functions
The impulse response function is a standard tool for understanding the properties of dy-
namic models. I define the impulse response function as the difference between the mean
of a variable following an impulse and the stationary mean of the variable. I character-
ize the impulse as a shift in the distribution relative to the stationary distribution. For the
exogenous spending impulse, the alternative distribution is the one obtained by applying
19
0.0
0.2
0.4
0.6
0.8
1.0
1.2
0.115 0.120 0.125 0.130 0.135 0.140 0.145 0.150
Capacity/capital ratio
Pre
se
nt
Va
lue
of
Wa
ge
Employer's Reservation Wage
Wage
Job-Seeker's Reservation Wage
Distribution
Figure 4. Reservation and actual wages as functions of z, for middle level of spend-ing
20
a special transition matrix to the stationary distribution. The special matrix alters the ex-
ogenous transition matrixπi,i′ by subtracting a small number from a diagonal element and
adding the same number to the element to the right. The effect is an upward shift in the dis-
crete stateg. For the technology shock, the alternative distribution is the density evaluated
at z − .01. The effect is to shift the capacity/capital ratio upward by 0.01.
I multiply the discretized stationary joint distribution ofz andg by the special transition
matrix to find the distribution of the two variables immediately after the impulse. I then
multiply that distribution by the standard transition matrix to find the distribution in the
period following the impulse. I continue the process for 20 quarters. From the distribution
of z andg in each period, I calculate the distributions of the other endogenous variables
and calculate their means.
Figure 5 shows the responses of unemployment over time to a positive productivity im-
pulse, for the flexible-wage and sticky-wage models. The scale is omitted from the vertical
axis, as the size of the impulse is arbitrary and the purpose of this and other similar figures
is to compare responses of different models to the same impulse. In both cases, unemploy-
ment falls immediately and gradually returns to normal (shown as a fine horizontal line).
The magnitude of the effect is far greater for the sticky-wage case. With flexible wages, the
incentive for job-creation,Z −W , hardly improves whenZ rises, becauseW rises almost
as much. With sticky wages,W rises as well, but not as much, so the incentive increases,
spending on recruiting rises, and the labor market tightens. In both cases, the dynamics
of z control the return to normal. The immediate effect of the technology impulse is a
jump inz—the impulse raises the numerator ofz while leaving its denominator, the capital
stock, unchanged. As capital accumulates following the shock,z returns to normal and all
variables, including the unemployment rate, also return to normal.
Figure 6 shows the responses of unemployment to an increase in exogenous spending.
Again, the response is much more dramatic for the sticky-wage model. In that model, the
initial effect of the impulse is to increase unemployment, because the increase in spending
21
0 4 8 12 16
Quarter
Un
em
plo
ym
en
t
Sticky wage
Flexible wage
0
Figure 5. Response of unemployment to a productivity shock
lowers the present value of future productivity,Z, immediately because it raises the ex-
pected future discount rate. This effect would be even stronger if labor supply were some-
what elastic. I explore these discount-rate effects below. After three quarters, a stronger
effect from risingz overcomes the initial effect and unemployment begins to fall below its
normal level. Eventually, unemployment turns around and returns to its normal level. The
process is prolonged because the spending shock is highly persistent.
5.3 The role of variations in the discount rate
One of the important contributions of general-equilibrium analysis of matching models of
the labor market is to assess the role of variations in the discount rate. Recall that the
present value of a worker’s productivity over the duration of a job, measured as of the
beginning of the job, is
Zt = γzt(1− ut)γ−1 + Pt ((1− st+1)Zt+1) xt. (33)
22
0 4 8 12 16
Quarter
Un
em
plo
ym
en
t
Sticky wage
Flexible wage
0
Figure 6. Response of unemployment to a spending shock
A higher discount rate lowers the second term on the right-hand side. The equilibrium
value ofZ is sensitive to the future discount rate over the roughly three-year period that the
job will last.
To measure the role of variations in the discount rate, I create a version of the model
in which the discount applied to the three labor-market variables,Z, U , andV , is constant
rather than endogenous. I fix the marginal rate of substitutionm at its stationary value from
the calibration. In this model, unemployment is substantially more volatile—the standard
deviation of unemployment is 3.9 percent, compared to 1.6 percent in actuality and in the
base case.
Endogenous discounting reduces the volatility of unemployment. When a positive pro-
ductivity shock hits the economy, the return to capital rises immediately. Consumption
growth rises to satisfy the consumption Euler equation. Both immediate and future dis-
count rates are higher. The value ofZ is correspondingly lower than it would be with a
23
constant discount rate. The reward to recruiting,Z − W , is also lower than it would be
with a constant rate. This factor offsets the direct stimulus from higher productivity in
raisingZ, which therefore rises less than it would with a constant discount rate.
In the case of a positive shock to exogenous spending, the reason for attenuation of the
effect on unemployment through the endogenous discount rate is similar, but not quite the
same. Because the spending shock has no effect on productivity and no initial effect on the
capital stock, the effects of the discount rate arise entirely through expectations of a higher
future discount rate. The spending shock results in a decline in investment, which drives
up the return as the shortfall in capital cumulates. The effect that is immediate in the case
of a productivity shock occurs over time for a spending shock.
5.4 Contributions of the two driving forces
Table 4 compares the standard deviations of the key variables in the full model (corre-
sponding to the first column of results in Table 3) and in structurally identical models that
have only the technology shock and only the exogenous spending shock. I find the exact
solutions to the modified models, so that these results apply to economies where partici-
pants know about the nature of the shocks. The left column of results repeats the standard
deviations from Table 3 for the full model. The middle column shows the effects of the
technology shock alone by setting the exogenous spending shock to zero. The right-hand
column shows the effect of the spending shock alone.
The spending shock is a little more important than the productivity shock for the invest-
ment/capital ratio, while the technology shock is a little more important for consumption
growth and the capacity/capital ratio and rather more important for unemployment.
6 Credible Threats in Wage Bargaining
Hall and Milgrom (2005) point out that the standard model in the Mortensen-Pissarides
tradition imputes threat points in the wage bargaining process that are not credible. A
24
Standard deviations Actual Base
Produc-
tivity
shock
alone
Exogenous
spending
shock
alone
Consumption growth 0.88 0.94 0.73 0.60
Investment/capital ratio 0.22 0.22 0.14 0.17
Log of capacity/capital ratio 4.43 4.59 3.66 2.37
Unemployment 1.57 1.59 1.34 0.73
Table 4. Contributions of technology and spending shocks
job-seeker and an employer, having found each other, enjoy a valuable joint surplus that
would be dissipated if one of the parties walked away from bargaining. Rather than walk
away, a party can take a step to create a successful match. The sequential-offer bargaining
model of Binmore, Rubinstein and Wolinsky (1986) provides a coherent framework for
studying credible threats. At each step in bargaining, one party may accept an offer or
make a counter-offer. The choice involves balancing the possibility of improving the terms
of the bargain by making a counter-offer and the cost of the delay resulting from making
the counter-offer rather than accepting the pending offer.
In the simplest model of the wage bargain based on alternating offers and credible
threats, the wage is
Wt =1
2
(bt
rt
− Vt + Zt
)(34)
Hereb is the net joint bias in the job-seeker’s favor resulting from delay—the sum of the
job-seeker’s benefit from spending time bargaining rather than working and the employer’s
cost of extending the bargaining process. During bargaining delay, the job-seeker avoids
burdensome work and continues to draw unemployment benefits. The employer incurs
costs of time devoted to the wage negotiations. The variabler denotes the short-term
25
discount rate applicable to the bias. Notice that the capitalized valuebt/rt replacesUt in
the earlier flexible-wage specification based on the Nash bargain with non-credible threat
points. Although the wage is disconnected from current conditions in the labor market as
measured byUt, it remains sensitive to the value of the rest of the job-seeker’s career,Vt.
This value influences the wage bargain because one of the consequences of delay in making
the bargain is to delay the receipt ofVt, which occurs at the moment the job-seeker makes
the bargain and begins work.
The model with completely isolated bargaining has unrealistic implications for volatil-
ity. Hall and Milgrom (2005) present a version of the model in which bargaining is less
isolated from conditions in the market because there is a hazard,µ, that the productive
opportunity will disappear during the bargaining process. The unique equilibrium in the
bargaining game becomes:
Wt =1
2
(bt
rt + µ+ Zt +
µ
rt + µUt − Vt
)(35)
In the GE model incorporating credible threats, I pick the parameterµ to match the
observed volatility of unemployment approximately. The value is 1.3, corresponding to a
likelihood of 0.5 that the bargaining process would end after 2.3 weeks, should the job-
seeker and employer fail to reach a bargain. In the equilibrium of the bargaining game, the
parties make a bargain instantly, so the possibility of bargaining for a period followed by
departure of the job-seeker to another employer never occurs.
In the stochastic growth model of this paper, the bargaining biasbt might reasonably
be constant in relation to the capital stock or constant in relation to the marginal product
of labor. I split the difference by making it a Cobb-Douglas combination of the two, with
equal elasticities of 0.5. I pick the value ofb so that, in the non-stochastic steady state,
the wage is the same in this model as in the base case. I also need to extend the model
by adding the short-term interest ratert as another endogenous variable. See Hall and
Milgrom (2005) for more details on these points.
26
Standard deviations Actual
Base--
sticky-
wage
model
Credible
threats in
wage
bargaining
Consumption growth 0.88 0.94 1.00
Investment/capital ratio 0.22 0.22 0.22
Log of capacity/capital ratio 4.43 4.59 4.83
Unemployment 1.57 1.59 1.54
Table 5. Volatilities, base model and credible-bargaining model
Table 5 shows the volatilities implied by the model with credible threats, withµ =
1.3. The model matches the observed values about as closely as does the base model with
sticky wages. Figures 7 and 8 show the responses of unemployment to productivity and
spending impulses when equation (35) governs wages, withµ = 1.3, in comparison to the
response for the earlier sticky-wage model. The credible-bargaining model responds much
less to productivity and much more to spending. The reason for the smaller response to
productivity is easy to explain—the value ofµ implies a tighter connection between labor-
market conditions and the wage. In equation (35), the coefficient onU is 1.30.025+1.3
= 0.990.
With respect to fluctuations in productivity, the credible-bargaining model almost has the
same amount of wage flexibility as the Nash-bargain model.
To explain the effect of higher exogenous spending, I examine the share of the surplus
accruing to employers:
Zt −Wt =1
2
(Zt − bt
rt + µ− µ
rt + µUt + Vt
). (36)
This quantity determines employer recruiting effort. Higher interest rates reduceZt +
Vt and thus reduceZt − Wt, reduce recruiting effort, and raise unemployment. But this
effect is offset by similar movements inUt in the opposite direction and the coefficient on
27
0 4 8 12 16Quarter
Un
em
plo
ym
en
t
Sticky wage
Credible bargaining
0
Figure 7. Response of unemployment to productivity impulse, sticky-wage andcredible bargaining models
0 4 8 12 16Quarter
Un
em
plo
ym
en
t
Sticky wage
Credible bargaining
0
Figure 8. Response of unemployment to spending impulse, sticky-wage and cred-ible bargaining models
28
Ut, as noted earlier, is close to one. The positive effect of the higher interest rate in the
denominator of the coefficients multiplyingbt andUt dominates. This effect is surprisingly
powerful.
In the credible-bargaining framework, fluctuations in unemployment arise primarily
from changes in exogenous spending and only secondarily from fluctuations in produc-
tivity. Traditional Keynesian macroeconomics had the same perspective. Although the
credible-bargaining model has the traditional property, the mechanism is rather different.
When high exogenous product demand drives up short-term rates temporarily, job-seekers
are at a disadvantage compared to normal times because extending bargaining is more
costly. The wage bargain shifts in favor of employers, who respond by recruiting more
aggressively and lowering the unemployment rate.
7 Fluctuations in Matching Efficiency
In the standard model, an exogenous decline in matching efficiency, modeled as a lower
value of the parameterω in equation (31), triggers a response in the labor market that
resembles a recession. Absent an explanation for such a decline in matching efficiency,
however, this account of a recession has little interest. Hall (2005a) describes a model
that generates fluctuations of this kind, based on changes in the extent of self-selection
among job applicants. When self-selection is high, matching efficiency is also high because
employers enjoy a high yield from evaluating applicants.
In the model, a job-seeker has private information about her probability of qualifying
for a particular job opening. A job-seeker sets a cutoff probabilityp∗ and incurs an appli-
cation costkW to apply for a job where the perceived probability of acceptance meets the
cutoff. The employer incurs an evaluation costkE and evaluates all applicants as long as
the expected payoff from a hire covers that cost. The parties split the surplus equally—the
model embodies the standard Nash bargain based on non-credible threats to disclaim the
potential relationship.
29
The expected probability that an applicant is qualified—given that applicants use their
private information and do not apply unless they know the probability to be at leastp∗—is
a functionχ(p∗). Job-seekers setp∗ according to
kW = p∗1
2S, (37)
whereS = Z − U + V is the surplus. Employers adjust their job postings to satisfy the
zero-profit condition,
kE = χ(p∗)1
2S. (38)
The ratio of the two defines a simple condition determining the cutoff:
χ(p∗)p∗
=kE
kW
. (39)
Changes in the distribution of the information available to job-seekers and changes in the
costskW andkE cause shifts in the cutoff probabilityp∗. The ratio of the costs is unlikely
to be an important driving force. But movements inp∗ resulting from changes in infor-
mation may be a potent driving force. When job-seekers become better informed,p∗ rises
dramatically.
Fluctuations arising from self-selection operate as follows: An event occurs, such as a
reallocation of labor from a shrinking sector to growing sectors. In consequence, the typical
job-seeker is less well informed about about the likelihood of qualifying for a given job
opening. Job-seekers lower their cutoff probabilities. Employers perceive a lowering in the
fraction of applicants who are qualified and reduce their recruiting efforts correspondingly.
Job-seekers respond by further decreasing their application cutoffs. The positive feedback
is less than complete and the labor market reaches a new equilibrium because the surplus
rises as the market slackens—job-seekers’ opportunity costsU − V are lower with higher
unemployment and the surplus is thus higher.
The stochastic GE framework of this paper cannot embed the full self-selection model.
Instead, I treat changes in the cutoff probability as generating changes in the matching-
efficiency parameterω. I double the number of discrete states indexed byg to include two
30
Standard deviations Actual Base
Flexible
wages and
variations in
matching
efficiency
Consumption growth 0.88 0.94 0.98
Investment/capital ratio 0.22 0.22 0.23
Log of capacity/capital ratio 4.43 4.59 4.87
Unemployment 1.57 1.59 1.63
Table 6. Volatilities with variation in matching efficiency
independent levels ofω. In the low-information state, I subtract a constant fromω and in the
high-information state, I add the same constant. In the results reported below, the constant
is 1.7. Thusω moves back and forth from 7.2 to 10.6 instead of holding at its constant value
of 8.9, as in the other models in this paper. I choose the switching probability to match the
observed persistence of unemployment.
Table 6 compares the volatility of an economy with variations in matching efficiency to
the volatilities of the actual U.S. economy and the base version of the model from earlier
in the paper. The model with flexible wages and variations in matching efficiency is about
as successful as the base model in matching the volatilities of the key macro variables.
8 Concluding Remarks
The equilibrium sticky-wage model, when incorporated in a complete dynamic stochastic
macro model, gives a reasonable account of the volatility of key macro variables, including
particularly the unemployment rate. A similar model with flexible wages set by a Nash
bargain falls far short of matching the actual volatility of unemployment.
Not only does the model with equilibrium wage stickiness match the standard deviation
31
of unemployment, but it is capable of explaining rare episodes when the unemployment
rate exceeds 8 percent. As Figure 1 shows, such episodes occur about once every 40 years.
They are the result of an unlucky succession of productivity disappointments that result in
a capacity/capital ratio that is well below its normal level. Wages fail to adjust fully, and
the economy finds itself far to the left in Figure 4, with little of the anticipated surplus from
employment accruing to employers. They put correspondingly little effort into finding new
workers, the job-finding rate is low, and unemployment is high. The situation gradually
corrects itself, as the capacity/capital ratio moves back to normal.
An alternative model of wage determination based on credible threats and sequential
offers yields wage stickiness and unemployment volatility similar to that found with the
equilibrium sticky-wage specification. The alternative model supplies an answer to the
primary objection to the equilibrium sticky-wage model, that it describes only an indeter-
minate equilibrium. The outcome of sequential bargaining is unique.
A third model capable of explaining observed unemployment volatility adopts the stan-
dard flexible-wage specification, but invokes shifts in matching efficiency. These shifts
may be the result of changes in the amount of information job-seekers have about their
likelihood of qualifying for jobs. Well-informed job-seekers self-select, thereby raising the
yield from employers’ recruiting efforts and raising matching efficiency.
The results in this paper suggest that modern ideas about friction in the labor market can
explain the volatility of employment and unemployment. Extensions of this investigation
would look at properties of the variables beyond their standard deviations—covariances
across variables and over time. Further, a full view of labor-market volatility probably
involves a mixture of the mechanisms explored here and others as well—see Hall (2005d)
for a review of other research with this goal. Mechanisms involving nominal frictions may
also prove important in a full explanation of the movements of unemployment. Further, I
have considered only a narrow range of driving forces. Monetary and other shocks belong
in a fuller model.
32
References
Alexopoulos, Michelle, “Unemployment and the Business Cycle,”Journal of Monetary
Economics, 2004,51 (2), 277–298.
Andolfatto, David, “Business Cycles and Labor-Market Search,”American Economic Re-
view, 1996,86 (1), 112–132.
Binmore, Ken, Ariel Rubinstein, and Asher Wolinsky, “The Nash Bargaining Solution
in Economic Modeling,”RAND Journal of Economics, Summer 1986,17 (2), pp.
176–188.
Hall, Robert E., “Intertemporal Substitution in Consumption,”Journal of Political Econ-
omy, April 1988,96 (2), 339–357.
, “Measuring Factor Adjustment Costs,”Quarterly Journal of Economics, August
2004,119(3), 899–927.
, “The Amplification of Unemployment Fluctuations through Self-Selection,” March
2005. NBER Working Paper 11186.
, “Employment Efficiency and Sticky Wages: Evidence from Flows in the Labor
Market,” Review of Economics and Statistics, August 2005,87 (3), 397–407.
, “Employment Fluctuations with Equilibrium Wage Stickiness,”American Economic
Review, March 2005,95 (1), 50–65.
, “Job Loss, Job Finding, and Unemployment in the U.S. Economy over the Past Fifty
Years,”NBER Macroeconomics Annual, 2005. Forthcoming.
and Paul R. Milgrom, “The Limited Influence of Unemployment on the Wage Bar-
gain,” April 2005. NBER Working Paper 11245.
33
Judd, Kenneth,Numerical Methods in Economics, Cambridge: MIT Press, 1998.
Lucas, Deborah, “Asset Pricing with Undiversifiable Risk and Short Sales Constraints:
Deepening the Equity Premium Puzzle,”Journal of Monetary Economics, 1994,34
(3), 325–342.
Lucas, Robert E., “Asset Prices in an Exchange Economy,”Econometrica, November
1978,46 (6), 1429–1445.
Merz, Monika, “Search in the Labor Market and the Real Business Cycle,”Journal of
Monetary Economics, 1995,36 (2), 269–300.
Mortensen, Dale T. and Christopher Pissarides, “Job Creation and Job Destruction in the
Theory of Unemployment,”Review of Economic Studies, July 1994,61 (3), pp. 397–
415.
Shimer, Robert, “The Cyclical Behavior of Equilibrium Unemployment and Vacancies,”
American Economic Review, 2005,95 (1), 24–49.
Summers, Lawerence H., “Some Skeptical Observations on Real Business Cycle Theory,”
Quaterly Review, Fall 1986,10, 21–27. Federal Reserve Bank of Minneapolis.
34
Appendix: Computing the Solution
I express the model as functional equations in the state variables,z andg, and their fu-
ture values,z′ andg′. Thuscg(z) is consumption when the current draw of the spending
shock isg and the output/capacity ratio isz. The spending shock takes onM possible
values. The future value ofz depends on the next draw of the productivity shock,a′:
z′g,a′ = a′xg(z)−γz. (40)
The future value of a variable such as consumption iscg′(z′g,a′). To avoid cluttered notation,
I will write z′ for z′g,a′ when the meaning is clear. The present value operator is
Pg(X) =∑
a′
∑
g′ψa′πg,g′mg,g′(z, z
′)Xg,g′(z, z′). (41)
The marginal rate of substitution is
mg,g′(z, z′) = β
(xg(z)
cg′(z′)
cg(z)
)− 1σ
. (42)
The return is
Rg,g′(z, z′) =
(1− δ)qg′(z′) + (1− γ)z′(1− ug′(z
′))γ
qg(z). (43)
The core functions in the model arexg(z), vg(z), Ug(z), Vg(z), andZg(z). The condi-
tions defining these functions are:
Pg (Rg,g′(z, z′)) = 1 (44)
Ug(z) = λγz(1− ug(z))γ−1 + Pg(Wg(z′) + Vg′(z
′))xg(z) (45)
Vg(z) = Pg (sg′(z′)fg′(z
′)(Wg′(z′) + Vg′(z
′)) + sg′(z′)(1− fg′(z
′))Ug′(z′) + (1− sg′(z
′))Vg′(z′)) xg(z)
(46)
Zg(z) = γz(1− ug(z))γ−1 + Pg ((1− sg′(z′))Zg′(z
′)) xg(z) (47)
35
vg(z) = (Zg(z)−Wg(z))sfg(z). (48)
The other equations of the model are:
Wg(z) = κW ∗ + (1− κ)1
2(Ug(z)− Vg(z) + Zg(z)) (49)
cg(z) =1− δ + z (1− ug(z))γ − 1−τ
2τ(xg(z)− 1)2 − vg(z)− xg(z)
1 + g(50)
qg(z) = 1 +1− τ
τ(xg(z)− 1) (51)
fg(z) = φ(vg(z)) (52)
sg(z) =s
1− ug(z)(53)
ug(z) = s (1− fg(z)) . (54)
The model has 11 endogenous variables,x, U , V , Z, v,W , c, q, f , s, andu. I associate
each with an equation, starting withx and equation (44) and ending withu and equation
(54).
I represent the core functions as Tchebysheff polynomials, as Judd recommends. Let
[a, b] include the support ofz and leth(z) = (2z − a − b)/(b − a), a function that maps
[a, b] into [−1, 1]. The polynomials are
T0(h(z)) = 1 (55)
T1(h(z)) = h(z) (56)
Ti+1(h(z)) = 2h(z)Ti(h(z))− Ti−1(h(z)) (57)
I represent the functionsxg(z), vg(z), Ug(z), Vg(z), andZg(z) as polynomials of orderN ,
such as
xg(z) =N∑
j=0
θx,g,jTj(h(z)). (58)
For L equally spaced values ofz in the interval[a, b], with L substantially larger thanN ,
the sum of squares of the values of the vector of conditions, equations (44) through (48),
36
-0.0025
-0.0020
-0.0015
-0.0010
-0.0005
0.0000
0.0005
0.00100
.10
24
0.1
06
4
0.1
10
4
0.1
14
4
0.1
18
4
0.1
22
4
0.1
26
4
0.1
30
4
0.1
34
4
0.1
38
4
0.1
42
4
0.1
46
4
0.1
50
4
0.1
54
4
0.1
58
4
0.1
62
4
Figure A. Solution errors
over the 5 equations,L evaluation points, andM discrete states, defines a norm of the
departures of the current values of theθ coefficients from representing a solution to the
model. I find the values of the5MN θ coefficients that minimize the norm. I substitute out
other functions such asug(z), using the other equations of the model.
To verify the accuracy of the solution and the polynomial approximations, I calculate
the values of the 5M-vector at 301 equally spaced points in the support ofz. Figure A
shows the values for the third equation, the one governingU .
To find the distribution ofz, I use 301 equally spaced bins spanning[a, b]. This creates a
discrete state variable with301M values. I compute the complete transition matrix implied
by the model from each of the values to each of the other values. I then solve for the
stationary probabilities by matrix inversion. I compute the impulse response functions
from the same transition matrix.
37